A New Support Vector Machine Algorithm with Scalable Penalty Coefficients for Training Samples

Abstract

In this paper, a new method to determine the penalty coefficients for different samples for the support vector machine (SVM) algorithm was proposed. Sequential minimal optimization (SMO) was then used to solve the SVM problem. Simulation results from applying the proposed method to binary classification problems show that the generalization error of the proposed method was smaller than standard SVM algorithm in the cases that the sizes of binary sample training sets (1) were selected in proportion; (2) were the same; (3) were quite different.